Skip to content

v1.6.0

Choose a tag to compare

@github-actions github-actions released this 09 Nov 09:11
· 17 commits to main since this release
ad524bc

A command given --server or --all could act on the wrong machine β€” install jellyfin --server rpi5 installed jellyfin here and docker restart nginx --all restarted it on every Pi β€” and deploy, alerts --all and a build for Windows could report a success or a reading nobody had measured; each now refuses, fails, or reads the real number.

Three things to check before upgrading. A script or MCP client config that passes --server or --all to a command that cannot use them, such as inventory or install, now gets an error instead of a local answer. --all with a command that changes a machine, such as docker restart, backup or wake, is refused; name one server with --server. And deploy and upgrade exit 1 when any server failed, which includes upgrade from a dev build. No Windows archive is published yet, so the Windows readings matter only to someone building homebutler for Windows themselves.

⚠️ Behavior changes

  • Commands that cannot reach another server refuse --server and --all instead of running here (#388). The two flags are global, and only the commands built to route remotely ever looked at them, so install jellyfin --server rpi5 installed jellyfin on this machine and inventory scan --server rpi5 --json answered with this machine's server_name. install and its subcommands, inventory, config validate, alerts init, alerts history, alerts test-notify, notify test, watch, serve, mcp, init, trust, proxmox and version now exit 1 before doing anything, naming the command and the flag: install runs on this machine only; --server is not supported. A script or MCP client config that passed either flag to one of them, and got a local answer, now gets that error; drop the flag. upgrade --server is refused too, because it upgraded every remote server while naming one; upgrade --all still works. backup drill --all, which drills every app in one archive, is drill's own flag and is unaffected. Nothing changes for the commands that already honoured the flags, or in any JSON output
  • --all refuses the commands it cannot summarise (#386). docker restart, docker stop, backup, restore and wake with --all were refused by the local server and sent to every SSH server, so docker restart nginx --all restarted it on each Pi while printing (parse error) for them. docker logs, docker top and docker inspect went the same way, and alerts --watch --all hung waiting on SSH servers that never answer, because a watch never returns. They now fail before any server is asked, with the list of what --all runs. Name one server with --server instead
  • deploy and upgrade exit non-zero when any server failed (#391). Both printed a βœ— for a server they could not reach or install on and then exited 0, so a script that deployed and went on to use the server never found out. Every server is still tried and printed, and the --json output is unchanged; the command then fails with deploy failed on 1 of 2 servers: rpi5 or upgrade failed on 1 of 3 targets: rpi5. upgrade counts the local machine as a target, so upgrade from a dev build, which refuses to replace itself, now exits 1. A binary deploy uploaded that then failed to run on the server was printed with a βœ“; it is now a βœ— and a failure
  • processes, report, docker stats, backup list and network scan now answer for a local: true server under --all, where it said not supported with --all (#386). report --all therefore saves a snapshot on the local server too, as it already did on each SSH one; --no-save compares without saving
  • GET /api/ports answers a failed read with a 500 and the reason, not a 200 with [] (#384). On Windows, or on a Linux host without ss, the dashboard's ports card said "No open ports detected" about a machine nobody had looked at. It now shows the collector's error, as the processes card already did. A successful answer is unchanged
  • the local server's alerts --all reading uses the alerts: thresholds in the config, as alerts does, rather than a fixed 90/85/90 (#386). The threshold in its --json --all entry changes for anyone who set their own. Its ports entry now names the container publishing a port, as ports --json does
  • doctor --strict --all exits non-zero when any server did not pass, naming them, including one that could not be reached or read (#386). --strict is no longer sent to each server, where a warning made that server's line remote command failed with its JSON dumped into the error
  • status and everything built on it refuse a platform they cannot read (#384). Outside Linux, macOS and Windows, status, alerts, report, doctor, inventory, the TUI and the dashboard's status answered with 0% CPU, 0 GB of memory and no disks; they now fail with unsupported OS: <name>, the error ports gives. No release is built for such a platform

πŸ› Fixes

  • --all without --json printed (parse error) for every SSH server, because the remote was asked for text and the line was read as JSON (#386). Each server is now asked for JSON whatever the user passed, and --json --all prints the same envelope as before
  • alerts --all showed a local: true server at 0% CPU, memory and disk, because every answer was printed as if it were status, while each SSH server printed (parse error) (#386). Each command --all runs now has its own line β€” alerts gives the worst level and the readings beside it, docker list counts running and stopped containers, report counts what needs attention and what changed β€” and a reading the answer did not carry is a -, never 0%
  • --all=true, --server=<name> and --config=<path> were passed on to each SSH server, which then fanned out again from its own config or looked for a server named after itself (#386). The = forms are now taken off like the spaced ones
  • deploy --local <path> with a path that does not exist connected to the server and could create ~/.local/bin there before finding out (#391). The file is now read before the server is contacted
  • status built for Windows printed 0.0% CPU, 0.0 / 0.0 GB of memory, no disks and an unknown uptime, and alerts, report, doctor, inventory_scan and the dashboard passed those on as a healthy, idle machine (#384). It now reads them from Windows: memory from GlobalMemoryStatusEx, CPU from GetSystemTimes over the same one-second window Linux uses, uptime from GetTickCount64, and each fixed drive from GetDiskFreeSpaceEx, reported with its root, such as C:\, as the mount. A subst letter is left out rather than listed as a second disk with the numbers of the drive it points into, which would also have alerted once per letter on one full disk. ports and processes still refuse on Windows, as they did
  • a server's key: ~\.ssh\id_ed25519 passed config validate on Windows and then failed to load when connecting, because only ~/ was expanded (#384)

πŸ“š Documentation

  • docs/multi-server.md says which commands take --server and --all, what --all runs and prints for each, and that everything else refuses them (#386, #388)
  • docs/configuration.md says where homebutler looks for its config, keys and known_hosts on Windows, and how to quote a Windows path in YAML (#384)